Upload
doanxuyen
View
227
Download
1
Embed Size (px)
Citation preview
Capital Structure Under Collusion
Daniel Ferrés
Universidad de Montevideo
Gaizka Ormazabal
IESE Business School
Paul Povel
University of Houston
Giorgo Sertsios
Universidad de los Andes
October 2016
______________________________ Gaizka Ormazabal acknowledges funding from the Ramon y Cajal and Marie Curie Fellowships and the Spanish
Ministry of Science and Innovation grant ECO2015-63711-P. We thank Murillo Campello, Francesco D’Acunto,
Mike Faulkender, Vojislav Maksimovic, Will Mullins, and Gordon Phillips for very helpful comments. We are also
indebted to seminar participants at the University of Maryland and University of Houston.
Capital Structure Under Collusion
Abstract
We study changes in financial leverage when firms actively collude
by forming a cartel. We find that cartel firms have lower leverage
ratios during collusion periods, which conflicts with the intuition of
the Trade-off Theory but is consistent with the intuition of
competition models based on trigger strategies. Cartels have a
surprisingly large economic footprint, because they tend to include
very large firms. Our findings thus help explain the puzzling
negative relation between profitability and financial leverage
reported in earlier work.
Keywords: Capital Structure; Financial Leverage; Financial
Policies; Collusion; Cartels; Trigger Strategies
JEL Classification: G32, L12
1
1. Introduction
We study the financial decisions of firms that join cartels, collusive agreements to raise
prices in the output markets. We find that during periods of collusion, cartel firms reduce their
leverage. This conflicts with the Trade-off Theory, according to which cartel firms should increase
their leverage, since price-fixing leads to increased profits, more valuable tax shields, and a
reduced threat of possible distress. Finding a negative relation between profitability and leverage
is actually common in the literature, and it has long been puzzling (Myers 1993; Parsons and
Titman 2008a; Graham and Leary 2011). We go beyond that, by studying one of the causes of
increased profits, cartel formation, and why this should relate to a reduction in leverage.
Understanding why cartel firms have lower leverage during collusion periods is important
for two reasons. First, it offers an explanation for the puzzling negative relation found in earlier
studies. Cartel firms represent a significant fraction of the economy, as they include very large
firms: Over the last two decades, publicly traded U.S. firms convicted of international cartel
activity accounted for more than one fifth of the total market capitalization (we describe the data
in more detail below). Second, our findings shed light on the role of financial strategies in the
functioning of a cartel. A more thorough understanding of how cartels operate is useful given the
deadweight efficiency losses that they cause (see Connor and Bolotova 2006).
Maksimovic (1988) offers a rationale for leverage reductions by colluding firms. The main
force in his model is the potential instability of collusive agreements. A cartel firm may have an
incentive to “cheat” by slightly undercutting the agreed-upon price, thus expropriating profits from
other cartel firms. One way to prevent this are so-called trigger strategies: Cartel firms interact
repeatedly, and if at any time a cartel firm deviates from the agreement, all firms switch to a
strategy of unfettered competition. If the threat is executed, all firms lose their share of the future
collusive profits. But since that also happens to the deviating firm, this threat can help prevent
deviations in equilibrium. Maksimovic (1988) adds a role for financial leverage: With high
leverage, a cartel firm has stronger incentives to deviate since the benefits from deviations are fully
internalized, but the costs from deviations are not (the potential loss is partially internalized by
creditors). Anticipating this, cartel firms looking for a stable collusive agreement should thus
ensure that no cartel firm has leverage that is too high. In other words, the cartel firms commit to
2
making their cartel stable by reducing their leverage. The prediction is thus that during collusion
periods, cartel firms should have reduced financial leverage. We refer to this as the “Commitment
Hypothesis” (see Section 2 for more details).
We study changes in leverage using a difference-in-difference approach. We compare
cartel firms to control firms, both during years in which cartel firms actively collude and years in
which they do not. Information about cartel membership for U.S. firms is gathered from the PIC
database, a comprehensive database on international price-fixing cartels (see Section 3 for details).
Consistent with the Commitment Hypothesis, we find that leverage is reduced by cartel firms
during periods in which cartel firms are actively colluding.
We corroborate our inferences using a series of triple-difference tests. We find that the
effect we document is concentrated among cartel firms with higher leverage (the need to reduce
leverage is stronger for them), cartel firms operating in more competitive environments (so
deviations are a larger threat, requiring larger reductions in leverage), cartel firms operating in
years of economic booms (when increases in current profits are more attractive and the threat of
reduced future profits has less bite), and cartel firms operating in environments with regulatory
developments aimed at destabilizing cartels (stronger reductions in leverage are then needed to
restore the stability of a cartel).
In additional tests we address the endogeneity of the timing of collusion. We instrument
the collusion and post-collusion periods, using the intensity of cartel activity or cartel dissolutions
in related industries. In the IV regressions we find an even larger reduction in leverage during
collusion. We also show that the changes in leverage by cartel firms are not caused by changes in
the cost of debt financing.
A comprehensive examination of a firm’s capital structure decisions requires analyzing all
major sources and uses of cash, including changes in payout, cash holdings, and investment. We
therefore study these broader financial policies of cartel firms. We find that cartel firms have
higher payout ratios during collusion periods, in the form of increased share repurchases.
Consistent with the finding in Hoberg et al. (2014), that firms facing stronger competitive pressure
tend to have lower payouts, we find that the payout increases are driven by cartel firms that face
less competitive environments (i.e., where cartels are more stable and the risk of the cartel breaking
3
up is likely lower). These are also the cartel firms that reduce their cash holdings during collusion
periods, consistent with cash holdings for precautionary reasons being less important in less
competitive environments (Haushalter et al. 2007).
In addition, we find that during collusion periods, cartel firms do not increase their capital
expenditures, but they do increase their R&D expenses. That firms do not increase their capital
expenditures during collusion is to be expected, as capacity expansions are counter-productive for
cartel firms. The increase in R&D expenses, on the other hand, is consistent with models that
predict that with reduced competition firms should invest more in R&D, since it is more likely that
they can internalize the benefits from such investments (Dasgupta and Stiglitz 1980; Phillips and
Zhdanov 2013). Taken together, our tests suggest that cartel firms make strategic changes to their
broadly defined financial strategies during collusion periods. This is important, as additional
evidence that firms strategically change their financial policies during collusive periods further
substantiates our inferences about the Commitment Hypothesis. In addition, we use these findings
to rule out possible alternative hypotheses that link financial leverage and profitability (see Section
6, for details).1
Our paper makes two important contributions. First, our results help understand the
negative empirical relation between profitability and financial leverage documented in the prior
literature. We show that it is important to analyze what drives changes in profits, because different
drivers have opposite effects on how leverage changes. There are, of course, many drivers. The
benefit of our focus on cartel membership is that it represents a very direct, positive shock to a
firm’s profitability, rooted in anti-competitive behavior that goes beyond what accounting
measures of profitability can capture. An alternative approach is taken in Xu (2012), who studies
shocks to a firm’s competitive environment, specifically, changes in import restrictions. She finds
a positive relation between profitability and leverage (as predicted by the Trade-off Theory).2 The
findings in Xu (2012) are the opposite of ours; the differences are due to the focus on different
1 We show in Section 6 why several theories that link profitability with financial leverage cannot explain our results,
including the Pecking Order theory (Myers 1984), strategic debt (Brander and Lewis 1986), growth options
(Strebulaev 2007), and the dynamic trade-off theory (Hennessy and Whited 2005). 2 A different approach is to consider differences in competitiveness across industries. MacKay and Phillips (2005)
find that firms in more concentrated industries have lower leverage ratios. The Commitment Hypothesis makes
predictions about within-firm changes in capital structure, not about cross-industry differences.
4
drivers of profitability, which confirms our argument that it is important to identify what causes
changes in profitability.
Our results also show (and explain why) firms within an industry can have very different
capital structures. The desire to show commitment to a collusive agreement makes cartel firms
choose leverage ratios lower than they would otherwise. But firms in the same industry that are
not members of the cartel choose leverage ratios that are unconstrained by commitment concerns
(their leverage ratios do not decrease when other firms in the industry collude).
The second contribution of our paper is to shed light on the inner workings of cartels, in
particular the role of leverage reductions by cartel firms trying to add stability to their illegal
collusive agreements.3 Cartels are important organizations, and they deserve attention in areas
outside of industrial organization. The economic footprint of cartel firms is large, as cartel activity
includes many large firms. 4 Focusing on U.S. firms included in Compustat, and aggregating over
the years 1990 to 2010, firms convicted of membership in at least one international cartel represent
16.6% of the total in terms of assets, 18.5% in terms of sales, and 22.7% in terms in terms of
market cap. Even if we compare only the firm-years during which cartels were actively colluding,
the proportions are high: 6.0% of assets, 5.2% of sales, and 6.3% of market cap. The importance
of colluding firms is further magnified because according to the literature, many cartels remain
undetected by the authorities (see Connor 2011, 2014, and Bryant and Eckard 1991), and the PIC
data does not include cartels that were not international.5 In addition, the significance of our results
can have even larger implications as the effects modeled in Maksimovic (1988) do not require an
explicit collusive agreement. Trigger strategies and the corresponding reductions in leverage can
be an equilibrium in the form of “tacit collusion”, which is much harder for anti-trust authorities
to prosecute than explicit cartels. In sum, cartel firms represent a significant part of the economy,
and their changed financial policies are likely to affect empirical work that pools cartel firms with
non-cartel firms without distinguishing them.
3 For an overview of how cartels operate, see Harrington (2006) and Levenstein and Suslow (2011). 4 That is what the literature on cartels has tended to find, see Levenstein and Suslow (2006); for a rationale, see Bos
and Harrington (2010). 5 Collecting data on cartels is difficult, see Connor (2014). All recent studies we know of use PIC data, or smaller
subsets of international cartels.
5
Our paper is related to the literature that studies the interaction between product market
decisions and financial policies (see Riordan 2003 and Parsons and Titman 2008b for overviews).
A few recent contributions focus on cartels and changes in antitrust policies. Dasgupta and
Žaldokas (2016) test whether “strategic debt” (Brander and Lewis 1986) or the need for financial
flexibility due to a threat of predation (Bolton and Scharfstein 1990) better explain leverage
changes after changes in antitrust policies, but they do not study the role of capital structure in the
functioning of the cartel (i.e., the Commitment Hypothesis). Dong et al. (2016) study how changes
in antitrust policies affect profits and M&A activity, but they do not consider the effects on capital
structure. Finally, Artiga et al. (2013) and Campello et al. (2015) also investigate cartel
convictions, but from the perspective of corporate governance.
The remainder of the paper proceeds as follows. Section 2 discusses the main hypotheses
tested in our paper. Section 3 describes the data. Section 4 presents results of the tests of these
hypotheses. Section 5 presents results about how cartel formation affects a colluding firm’s more
broadly defined financial policies, including payout policy and cash holdings. Section 6 discusses
possible alternative explanations, and Section 7 concludes.
2. Hypotheses
Models of the Trade-off Theory are developed in Kraus and Litzenberger (1984) and
Bradley et al. (1984). The intuition for how profitability should affect leverage is straightforward
(see also Xu 2012): With higher profits, the threat of financial distress is reduced, and the tax
benefits of corporate debt become more significant. Aligned, the two forces should lead to higher
leverage. That should apply to cartel firms when collusion starts, since the goal of cartels is to
increase profits. So the Trade-off Theory Hypothesis predicts that when firms form a cartel, their
leverage should increase.
The opposite prediction is made by what we refer to as the “Commitment Hypothesis.”
Maksimovic (1988) studies how firms may be able to sustain high prices and low outputs without
the ability to legally enforce any such agreements (since they are illegal), and how this is affected
by financial leverage. Collusion is feasible in a multi-period model. If firms know that they will
6
interact repeatedly, they have an incentive to set the agreed high prices (and resist the temptation
to lower their prices for an instantaneous profit at the expense of the other cartel members) if such
deviations lead to costly punishment in later periods.
A credible threat is created by “trigger strategies”: Each firm plans to choose the collusion
action (high prices, low output) in each period, indefinitely, but if one or more rival firms deviate
and charge lower prices, all firms revert to a low-price equilibrium strategy in each period that
follows.6 That response is credible, and it is a punishment threat since it reduces future profits for
all firms, including firms that deviated. (This punishment outcome is costly for all firms, but in
equilibrium it can be avoided.) Collusion can be sustained if the collusion profits that are lost after
a deviation are more valuable than the one-time profit that can be earned by deviating in any period.
Note that an explicit collusive agreement is not needed: Such trigger strategies form an equilibrium
and can arise in the form of “tacit collusion”.
Maksimovic (1988) shows that financial leverage exacerbates the incentive to deviate from
collusion and must therefore be kept in check. With significant debt, a cartel firm’s future profits
largely go to the creditors, so the threat of lower future profits has less bite (the shareholders’
lowest possible future payoff is zero, due to limited liability); and the shareholders benefit if the
firm can earn a large instantaneous profit (net of debt payments) by deviating. If several firms
decide to form a cartel, they must therefore ensure that each cartel firm’s leverage is below a
certain level, thus strengthening its commitment to abide by the collusive agreement. The
Commitment Hypothesis thus predicts that when firms form a cartel, the average leverage ratio
should be reduced while profits increase.
3. Data and Variable Construction
Our analysis uses the Private International Cartels (PIC) database, which contains
information on virtually all private international price-fixing cartels detected by antitrust
authorities between 1990 and 2012.7 The database is described in detail in Connor (2014). The
focus is on “private” cartels, since government-sanctioned “public” cartels are not at risk of
6 The role of repeated interaction in making tacit collusion feasible was first described in Friedman (1971). 7 The data includes a few cartels discovered before 1990. Where possible, we use data from 1985 onwards in our tests.
7
prosecution; and the data include only cartels with an “international” flavor, i.e., cartels that
include firms from multiple countries, or if an antitrust authority pursued firms registered abroad.
The information in the PIC database is collected from press releases by antitrust authorities
such as the Department of Justice and the Federal Trade Commission in the U.S., the European
Commission (Directorate-General for Competition), or Canada’s Competition Bureau. Firms are
included in the database if an antitrust authority imposed fines or if class action lawsuits were filed.
Since many cartels remain undetected (Connor (2014) estimates that only about 10-30% of all
cartels are detected; see also Bryant and Eckard 1991), the data does not include all cartels but
only those that were detected and for which a conviction was possible.
From this database we collect the following information for each cartel firm: Name,
country of incorporation, markets and locations where collusion took place, and the start and end
dates of the collusive agreements. We restrict the sample to U.S. firms, since several of our tests
use additional data sets that focus on U.S. firms. We require that these firms are included in
Compustat, which is the case for 216 firms. We exclude firms involved in more than one cartel
simultaneously, since otherwise the start and end dates of collusion would be arbitrary. The final
sample includes 1,427 firm-years for 93 cartel firms.8
In several of our tests we include “non-cartel” firms as controls. These are U.S. firms
included in Compustat that were not cartel firms, i.e., they were not in the PIC database. We
exclude non-cartel firms that operate in the same 4-digit SIC code as a given cartel firm, since
Leary and Roberts (2014) show that firms imitate their rivals’ decisions to some extent. If so, the
tests could fail to detect leverage changes merely because same-industry non-cartel rivals changed
their leverage too. As a robustness check, we replicate our tests including same-industry rivals.
Our results continue to hold, suggesting that the pattern we document does not reflect a general
variation in industry conditions. In fact, we show below (Section 4.2) that these non-cartel firms
increase their leverage during collusion years, so including them in the control sample would make
it easier to find evidence that cartel firms decrease their leverage during collusion years.
8 The cartel firms in our sample belonged to 58 different cartels. The other cartel members included international
firms, as well as publicly traded and privately held U.S. firms.
8
The sample of non-cartel firms includes 127,826 firm-years for 12,963 firms. An
alternative use of control firms would be to construct matched samples. The drawback of such an
approach is that the most similar firms (in terms of observable characteristics) would also most
likely be undetected members of cartels (for example, size is a strong predictor of cartel
membership; see also Dong et al. 2016).
Table 1 presents descriptive statistics for selected variables. Panel A presents summary
statistics for the data set including both cartel and non-cartel firms. Panel B presents separately
the means for cartel and non-cartel firms, and the significance levels for the differences. The main
variables used in our analysis are Profitability, Leverage, Total Payout, Dividend and Cash. All
five are scaled by the book value of assets; for more details on the construction of each variable
see Appendix A. All financial variables from Compustat are winsorized at the 1% level.
Table 1
The statistics in Panel B of Table 1 show that cartel firms are larger and more profitable
than non-cartel firms. They have lower leverage and cash holdings than non-cartel firms. Their
overall payout to shareholders is higher, but that is driven by higher share repurchases — while
cartels firms are more likely to pay dividends, the average dividend (scaled by total assets) is
smaller. Finally, cartel firms exhibit low cash flow volatility and high asset tangibility.
How long a cartel is active varies across the sample: The average duration is well over six
years, and the median duration is just under five years; 12.7% lasted for less than a year, while
7.2% lasted 15 or more years (the maximum is 34 years). There is also variation in the number of
firms that joined a cartel. The average is less than seven, the median five; just over 15% of the
cartels consisted of only two firms, while 8.3% consisted of 15 or more firms (the maximum is 42
firms). These numbers are consistent with those in earlier studies (Levenstein and Suslow 2006).
4. Collusion and Financial Leverage
4.1 Empirical Design
To analyze the relation between collusion and firms’ financial policies we estimate
variations of the following baseline empirical model:
9
𝑦𝑖𝑡 = 𝛼 + 𝛽 ∗ 𝐶𝑜𝑙𝑙𝑢𝑠𝑖𝑜𝑛𝑖𝑡 + γ ∗ 𝑃𝑜𝑠𝐶𝑜𝑙𝑙𝑢𝑠𝑖𝑜𝑛𝑖𝑡 + 𝜴´𝑿𝒊𝒕−𝟏 + 𝜑𝑖 + µ𝑡 + 𝜀𝑖𝑡 (1)
The subscript i indexes firms, and t indexes years. Our main dependent variable, yit, is book
leverage. Our specification is essentially a difference-in-difference strategy with two treatments:
Collusion and Post Collusion. Collusion takes a value of 1 for cartel firms during cartel years, and
0 otherwise; Post Collusion takes a value of 1 for cartel firms during the 5 years after a cartel is
dissolved, and 0 otherwise. This research design compares differences between collusion years of
cartel firms with both non-cartel firms and non-collusion years of cartel firms. It also differentiates
years shortly after collusion from other non-collusion years to capture the potential effects of a
cartel’s dissolution on its members’ financial policies.
Our key parameter of interest is 𝛽. A positive estimated coefficient would be consistent
with cartel firms raising leverage ratios during cartel years, in response to increased profits; a
negative coefficient would be consistent with firms reducing leverage ratios during collusion years,
to strengthen their commitment not to deviate from a collusive agreement.
We include a set of controls X that includes variables commonly used in the capital
structure literature (e.g., Lemmon et al. 2008): Lagged tangibility, lagged profitability, lagged
sales, lagged median industry leverage and lagged cash flow volatility. Firm and year fixed effects
are represented by 𝜑𝑖 and µ𝑡, respectively.
In some specifications we use cartel firms only (i.e., the “eventually treated” sample). In
this setup, non-collusion firm-years are the control sample. To the extent that the results are similar
to those of the main specification, this specification helps to attribute changes in behavior to the
treated group, rather than to the control group. In all our specifications we adjust standard errors
for heteroscedasticity and industry clustering. We cluster standard errors at the industry level
because firms compete and collude at this level of aggregation. This clustering strategy allows for
three types of arbitrary correlations in the error term: (1) Error correlation across different firms in
a given industry and year; (2) error correlation across different firms in a given industry over time;
and (3) error correlation for a given firm over time (see Petersen (2009)).
10
4.2 Leverage During and After Collusion Years
Table 2, Panel A, presents the results for our baseline empirical specification, Equation (1).
Columns (1) and (2) show the results using both cartel and non-cartel firms. Columns (3) and (4)
show the results using cartel firms only (i.e., the “eventually treated” sample). In columns (1) and
(3) we present the results from regressions without controls, while in columns (2) and (4) we
control for capital structure determinants previously used in the literature. In all four regressions,
Collusion has a significant negative effect on leverage, reducing it by 2.5 to 3 percentage points.
These effects are statistically significant at the 5% level, and they are economically significant too:
As shown in Table 1, the average leverage ratio of cartel firms is 27%, so the leverage ratio
decreases by nearly a tenth. Furthermore, the effect is twice as large for high-leverage firms, as
we discuss below (see Table 3). The coefficients for Collusion are very similar across the four
columns, which suggests that the effect is driven by cartel firms reducing their leverage ratio, and
not by leverage changes made by non-cartel firms (recall that columns (3) and (4) use cartel firm
data only). Overall, the evidence supports the Commitment Hypothesis and is not consistent with
the intuition of the Trade-off Theory.
Table 2
Table 2, Panel A, also suggests that leverage is somewhat reduced in the years after
collusion ends (the coefficients on Post Collusion are negative, but they are not statistically
significant). This may capture increases in the cost of debt financing due to negative reputation
effects after having been convicted of cartel activities (see Section 4.6, where we show that cartel
firms face a more difficult credit environment after cartels are dissolved). It may also be a
reflection of a return to competition, leading to lower profits.
Table 2, Panel B, shows corresponding regression results for the firms in the same industry
as cartel firms, but not members of the cartel. That is, we study whether non-cartel firms also
reduced their leverage when their industry rivals formed a cartel. We find no evidence for that, so
our results are not driven by industry trends. In fact, in the regression with controls (column 2 in
Panel B of Table 2), the coefficients are positive (instead of negative) and statistically significant,
so the non-cartel firms increased their leverage during the collusion period. That could be
consistent with the Trade-off Theory, if those firms benefited indirectly from the cartel’s collusion:
11
Maybe the non-cartel firms were also able to charge higher prices and earn higher profits (see Bos
and Harrington 2010), and therefore it was optimal to increase leverage. The coefficient for the
post-collusion period is also positive, which is also consistent with the Trade-off Theory if the
non-cartel firms enjoyed an improved market position after the cartel was dissolved.
4.3 Can Leverage-Ratio Trends Explain the Results?
A possible concern with our estimates in Table 2, Panel A, is that the negative coefficients
for Collusion could reflect a time trend in the financial policies of cartel firms that is unrelated to
collusion but overlaps with the period of collusion. That could invalidate the “parallel trends”
assumption that underlies our difference-in-difference strategy. Nevertheless, we address this
concern empirically by modelling time trends in Leverage around the collusion period.
Specifically, we estimate the following variation of Equation (1):
𝑦𝑖𝑡 = 𝛼 + 𝛽 ∗ 𝐶𝑜𝑙𝑙𝑢𝑠𝑖𝑜𝑛𝑖𝑡 + ∑γℎ ∗ 𝑑𝑖ℎ + 𝜑𝑖 + µ𝑡 + 𝜀𝑖𝑡. (2)
The subscript h indexes the years that immediately precede collusion years (h {−3, −2, −1}),
years that immediately follow collusion years (h {1, 2, 3}), or years that are collusion years (h =
0 for full collusion years, h = 0– for a partial collusion year at the start of a cartel, and h = 0+ for a
partial collusion year at the end of a cartel). For example, if a cartel started in (say) September
2001 and ended in May 2005, then di0- = 1 in 2001 and zero otherwise; di0 = 1 in 2002, 2003 and
2004 and zero otherwise; di0+ = 1 in 2005 and zero otherwise; and so on. We distinguish full years
of collusion from partial years of collusion, since during partial years of collusion the effects are
likely weaker, so this more granular data analysis is more informative. Pooling partial and full
years yields very similar results.
We plot the regression coefficients γℎ and their 95% confidence intervals in Figure 1. In
Panel A, the coefficients are estimated using Leverage as the dependent variable. The results
suggest that the decrease in Leverage is concentrated in the collusion years. In the years preceding
or following collusion years, and in partial collusion years, the coefficients are not significantly
different from zero. Overall, this suggests that the empirical pattern we document in Table 2, Panel
A, is unlikely to be driven by time trends in the financial policies of firms that form cartels.
12
Figure 1
The goal of collusion is to increase the profits of the cartel firms. To confirm that profits
do indeed increase during collusion periods, we re-estimate Equation (2) using Profitability as
dependent variable, and we present the coefficients γℎ and their 95% confidence intervals in Panel
B of Figure 1. Profitability declines in the years before cartels are formed, in particular in year h
= –1, the year immediately preceding the formation of a cartel. This suggests that disappointing
profits could be a motivation for the formation of a cartel. Profitability starts to increase during
the first partial year of collusion (h = 0-), but the increased profitability is more apparent during
the full collusion years. Not surprisingly, Profitability drops substantially after the cartel is
dissolved and competition resumes.9
While Figure 1 shows that leverage drops during collusion years (and not before), it does
not show during which collusive years leverage falls. Understanding the exact timing of the drop
in leverage is important, as the Commitment Hypothesis predicts that leverage should be reduced
when the cartel is formed, in order to mitigate any risk of cartel members violating the agreement.
We explore the timing of the leverage changes in more detail in Figure 2. Panel A shows the
average leverage of cartel firms for the years preceding collusion and the first few years of
collusion, with year 0 being the year in which collusion started. There is a clear drop in leverage
during the first two collusion years, followed by a partial rebound. This is consistent with firms
reducing their leverage aggressively at the beginning, when a cartel is formed, and with drops in
this financial discipline as time proceeds, undermining the stability of the cartels and possibly
leading to their break-up. We find a similar pattern when running an extended version of equation
(2), in which we differentiate the effects on leverage for the first three full collusion years from
those of other collusion years. We display the coefficients in Figure A.2 in the appendix. The figure
shows a strong reduction in leverage for the first two cartel years and a partial rebound afterwards.
In Panel B of Figure 2 we examine in more detail how the change in leverage during the
first few collusion years happens, by displaying the evolution of both total debt and assets. The
9 To confirm that profits change as expected when cartel firms collude, we compute the correlation between the
profitability of the members of each cartel, and we display the averages of those correlations (for the three groups of
firm-years) in Figure A1, in the Appendix. The average correlation is higher for collusion years, and it is lower (in
fact, negative) for firm-years classified as post-collusion years.
13
figure shows that the reduction in leverage is driven by a strong reduction in debt, which is larger
than the reduction in assets observed for the same event years.
Figure 2
4.4 Specific Predictions of the Commitment Hypothesis: Triple-Differences Tests
We now present a battery of triple-differences tests aimed at sharpening identification.
These tests exploit cross-sectional and time-series variation in our data to further check whether
the decrease in leverage during collusion periods we document is consistent with the Commitment
Hypothesis.
4.4.1 Commitment is Easier for Low-Leverage Firms
The Commitment Hypothesis predicts that all else equal, cartel firms with relatively higher
leverage need more dramatic leverage reductions during collusion periods. To explore this
prediction, we compute each firm’s leverage ratio during the first year of data availability in our
sample, and then we split the sample in two: firms with high initial leverage (above the sample
median), and firms with low initial leverage (below the sample median). We then estimate
Equation (1) separately for the two subsamples. The goal is to distinguish firms that had high
leverage in pre-collusion years from firms had low leverage, and by using the first year of data
availability we can include the control firms in this test (for non-cartel firms, there is no well-
defined “pre-collusion” period). We use the first year of data availability, since financial leverage
seems to be “sticky” (Lemmon et al. 2008). We have repeated the same estimations both using
each firm’s average leverage during its first three years in the sample, and using the average of all
firm-years available for a given firm, with very similar results.
We present the results in columns (1) and (2) of Table 3. The decrease in leverage is
significant for high-leverage firms, but it is insignificant for low-leverage firms. The decrease in
leverage by high-leverage firms is economically large, at about six percentage points. That
reduction is twice as large as that reported in Table 2 for the pooled sample. This supports the
Commitment Hypothesis, as high-leverage firms strategically reduce their leverage in order to add
credibility to the collusive agreements, while low-leverage firms do not need large decreases in
leverage. The leverage levels remain below the pre-collusion levels in the years after collusion
14
ends for firms with high initial leverage. As we show below, the cost of debt financing increases
after cartels are dissolved, which explains this result.
Table 3
4.4.2 Commitment is Easier with Weaker Competition
The second test exploits variation in competitive pressure across firms. With stronger
competitive pressure, it is likely harder to sustain collusive agreements: The up-front profits from
deviating (undercutting the cartel prices) seem large, compared with competitive profits; and the
future losses due to any “punishment” by rival firms seem less threatening, in particular if
shareholders consider the benefits of limited liability in case the punishment phase turns out to
have severe consequences. The Commitment Hypothesis thus predicts that where firms face
stronger competitive pressure, a more significant reduction in leverage is required to sustain
collusive agreements.
Measuring the competitive pressure a firm faces is not straightforward, since there are
many ways to differentiate a firm’s products. A new measure is “product-market fluidity,” used
in Hoberg and Phillips (2010, 2016), Hoberg et al. (2014), and other papers. It uses textual analysis
of product descriptions found in SEC 10-K forms to estimate the intensity of a firm’s product-
market threats. The sample used in Hoberg et al. (2014) covers the years 1997 to 2008, so it does
not cover all years in our sample. However, fluidity seems to vary little over time, so we compute
each firm’s average fluidity and use those averages for our entire sample period. We split our
sample into observations with above-median fluidity (stronger competitive pressure) and below-
median fluidity (weaker competitive pressure) and then estimate Equation (1) separately for the
two subsamples.10
The results are presented in columns (3) and (4) of Table 3. We find that firms facing
stronger competitive pressure (high fluidity) reduce their leverage substantially when colluding,
while firms facing weaker competitive pressure (low fluidity) do not. Notice also that leverage
remains low (or even decreases further) after the cartels are dissolved. With a return to
10 The subsamples are not of equal size as we compute the median fluidity using the overall sample, including firm-
years with missing data needed in our regressions. We find similar results if we first eliminate observations with
missing data and then compute the sample median of fluidity to split the sample.
15
competition, it seems reasonable that firms facing stronger competitive pressure maintain low
leverage ratios: presumably, profits are lower, and a firm’s competitive situation is likely tougher.
This would be consistent with the Trade-off Theory.
4.4.3 Commitment Is Easier During Recessions
The third test exploits time-series variation in the incentive to deviate from a collusive
agreement. The incentive to deviate is stronger if the up-front benefits are larger and the profits
lost if collusion ends smaller. Rotemberg and Saloner (1986) predict that collusion is harder to
sustain during economic booms, when larger profits are generated in the aggregate and thus there
are larger potential gains from deviating. Analogously, they predict that it is easier to sustain
collusion during recessions, when the profits from any deviation are smaller. We use this intuition
as the basis for our next test: Cartel firms should decrease their leverage by more during boom
years, and by less during recession years.
Because this test is based on time-series variation in the data, we can also test whether our
earlier findings might be driven by cross-sectional unobserved heterogeneity. For instance, it could
be argued that cartel firms are more sensitive to changes in the competitive environment than non-
cartel firms, and this is why we observe decreases in leverage for the collusion and post-collusion
periods. And if highly levered cartels firms (of cartel firms in highly competitive environments)
are somehow even more sensitive to these changes, this could also explain the results from the
previous triple-difference analyses.
Specifically, in this test the post-collusion years are used as a placebo. There should not be
any significant differences in leverage in the post-collusion period for cartel firms, across recession
and non-recession years. If that is what we find, then it is unlikely that our results are simply due
to cartel firms being more sensitive to economic shocks. Furthermore, if we find differences in
leverage in the collusion period but not in the post-collusion period, this would add support to the
Commitment Hypothesis.
To test whether cartel firms decrease their leverage less during recession years, we conduct
a triple-difference test by interacting Recession Year, an indicator variable for recession years,
16
with both Collusion and Post-Collusion. Recession years are identified using NBER data. The
results are presented in Table 4, column (1).
As before, we find that there is a significant decrease in leverage during collusion years.
Importantly, macroeconomic conditions seem to have a significant impact on changes in leverage:
The interaction term of Recession Year and Collusion is positive. These coefficients all have
significant economic sizes. In sum, cartel firms decrease their leverage by less (or not at all) during
recession years, consistent with the intuition gained from Rotemberg and Saloner (1986).
The coefficient of the interaction between Recession year and Post Collusion is
insignificant. This suggests that cartel firms do not change their leverage differently when
colluding during recession years simply because they are more sensitive to shocks (or they are
somehow different in unobserved dimensions to non-cartel firms).
Table 4
4.4.4 Commitment is Easier in the Absence of Leniency Laws
The fourth test exploits time-series and cross-sectional variation in the stability of cartels,
based on differences in the rewards for disclosing cartel activities to the antitrust authorities.
Leniency laws are regarded as a significant threat to collusive agreements (see OECD 2002; Miller
2009; Dong et al. 2016; and Dasgupta and Žaldokas 2016). These laws offer immunity to the first
member of a cartel who confesses to the authorities. Other cartel members get at best partial relief
if they also confess and cooperate.11 This creates an incentive to be the first to betray a cartel to
the authorities, putting strain on a cartel’s ability to sustain a collusive agreement.
Leniency laws are a recent legal innovation, adopted in many countries starting effectively
in 1993. They have not been introduced in all countries, and the date of their passage varies across
countries, i.e., the U.S. firms in our sample faced leniency laws (and their destabilizing effect on
cartels) in some countries but not others, and the intensity of this exposure (the number of countries
with leniency laws) varied over time. Thus, there is both cross-sectional variation and time-series
variation in the exposure of cartels to leniency law regimes.
11 For details of leniency laws in various countries, see Lex Mundi (2013).
17
The passage of leniency laws in various countries offers us an additional identification
strategy. It is reasonable to assume that the passage of these laws is an exogenous event when
considering a given cartel: An as-yet undiscovered cartel is unlikely to have an effect on antitrust
laws. This staggered introduction of leniency laws is also used for the identification strategy in
Dasgupta and Žaldokas (2016) and Dong et al. (2016). The Commitment Hypothesis predicts that
if a country introduces or has leniency laws, it should be harder to sustain collusive agreements,
and thus stronger reductions in leverage ratios are needed to sustain collusion. The more countries
covered by the cartel that have leniency laws, the stronger this effect should be.
We restrict our sample to cartel firms since information about which firm operates in which
country is available only for cartel firms. We create a firm-year dummy High Leniency, which
takes a value of one if a cartel operates in three or more countries with leniency laws, and zero
otherwise. (The lowest number of countries is zero, the highest number is seven.)
In our triple-differences specification we interact the dummy High Leniency with Collusion
and Post Collusion. The Commitment Hypothesis predicts that the coefficient for Collusion
should be negative, and that the coefficient for the interaction term with High Leniency should also
be negative. There is no clear prediction for the coefficient of the interaction term of High
Leniency and Post Collusion. On the one hand, cartel firms shouldn’t change their leverage
differently in the post-collusion period, as the incentives to reduce leverage during collusion
periods are no longer in place. On the other hand, if a cartel is dissolved in a rigorous antitrust
environment (which is likely the case in countries that have passed leniency laws), this may have
a particularly strong effect on the firms’ reputation and their subsequent ability to raise funds (we
explore this in Section 4.6). Cartel firms in the post-collusion period may thus keep their leverage
low in High-Leniency environments.
We present the results in Table 4, column (2). As in the main specification, the coefficient
for Collusion is negative. Importantly, cartel firms more exposed to leniency laws exhibit more
pronounced decreases in leverage during collusion years: The interaction term of High Leniency
and Collusion is both economically and statistically significant. The results are thus consistent
with the predictions of the Commitment Hypothesis. The interaction term of High Leniency and
18
Post Collusion is negative, but only marginally statistically significant (at the 10% level). This is
to be expected given the conflicting effects described above.12
4.5 Endogeneity of the Collusion Periods
One possible concern is that whether a cartel is active or dissolved in a given year is itself
endogenous. That is, the formation of a new cartel may be driven by factors that also affect
leverage choices, without any direct causal effect from membership in a cartel on financial
leverage choices. While our triple-difference results mitigate concerns regarding omitted variables
or reverse causality, we now more directly address endogeneity concerns using an instrumental
variables approach.
We construct instruments based on environmental factors that facilitate or impede cartel
formation. The decision to form a cartel and the ability to sustain it depend on the power of antitrust
authorities, including the likelihood that a cartel is detected and the penalties that can be imposed.
Symeonidis (2003) finds evidence that the likelihood of collusion is industry-specific. Differences
in transparency and in the structure of competition across industries suggest that collusion may be
easier to sustain in some industries than in others (obtaining information from customers or
whistleblowers may be harder in concentrated industries, for example). Furthermore, there are
differences in the power and goals of antitrust authorities, even within countries: in the U.S.,
antitrust concerns may be raised by either the FTC or the DOJ, depending on the industry (in
addition, state-level authorities may initiate antitrust proceedings).
We construct two industry-level proxies for the probability of prosecution of a cartel in any
given year. First, for a given firm-year, we count the number of firms with the same 2-digit SIC
code that colluded during the same year or the preceding two years, excluding firms with the same
4-digit SIC code as the cartel firm. We denote this measure by Cartels Active. Firms with the
same 4-digit SIC code are excluded to avoid a simple mechanical correlation between the
instrument and the potentially endogenous regressor. Second, we construct a similar measure
counting firms whose cartels were dissolved during the same year or the preceding two years,
12 Only three firms in our cartel sample broke up their cartels by taking advantage of leniency deals. Our main results
are robust to excluding those firms.
19
again considering only firms with the same 2-digit SIC code but excluding firms with the same 4-
digit SIC code. We denote this measure by Cartels Dissolved.13
It seems plausible that a large number of prosecutions in a broadly defined industry means
that collusion is generally harder to sustain, i.e., that a higher value of Cartels Dissolved makes it
less likely that a cartel is formed. Similarly, it is plausible that the presence of many cartels in a
broadly defined industry means that collusion is easier to sustain in that group of industries, i.e.,
that a higher value of Cartels Active makes it more likely that a cartel is formed. Ideally, these
instruments could be augmented by using direct measures of the DOJ’s and FTC’s investigative
power, e.g., by using their annual budgets. However, being time series, such data would be
perfectly collinear with time fixed-effects, precluding their use in our difference-in-difference
setting.
In practice, large firms are more likely to join international cartels than small firms (Connor
2014; Dong et al. 2016). This allows us to refine our proxies for exogenous variation in the
probability of cartel detection and prosecution. We sort the firms by size and create size quartile
dummies that we then interact with the two prosecution-probability proxies, Cartels Active and
Cartels Dissolved.
Columns (1) and (2) of Table 5 present the results of the first stage of the IV estimation.
The results show that collusion and post-collusion periods are strongly associated with our
instruments for cartel formation and prosecution (F-test of 139). The signs of the coefficients are
as expected: Cartels Active is positively associated with Collusion and negatively associated with
Post Collusion; while Cartels Dissolved is negatively associated with Collusion and positively
associated with Post Collusion. Columns (3) and (4) of Table 5 present the corresponding results
allowing for interaction of the two instruments with size quartiles. We find that the two instruments
have a stronger effect for larger firms. Given this higher goodness of fit, we therefore use this
second specification as our first stage when estimating the second stage.14
13 The results are robust to considering firms with the same 3-digit SIC code but a different 4-digit SIC code; or firms
with the same 2-digit SIC code but a different 3-digit SIC code. 14 Our IV estimates are not sensitive to the number of size partitions that we interact with the main instruments. In the
appendix, Table A.1, we present second stage results that use the interactions of Cartels Active and Cartels Dissolved
20
The results from the estimation of the second stage are shown in column (5) of Table 5. As
before, we find that collusion has a negative and significant effect on financial leverage. The
magnitude of the coefficient on Collusion is larger than in the earlier OLS specifications. The IV
regressions thus further confirm our earlier findings, that cartel firms reduce their leverage ratios
during collusion periods, consistent with the Commitment Hypothesis and conflicting with what
the Trade-off Theory predicts. Overall, the evidence in Table 5 suggests that the potential
endogeneity of the start and end dates of a cartel is not driving our results.
Table 5
4.6. Collusion and the Cost of Debt Financing
A potential concern about our prior results is that the decreases in leverage we observe
could be due to contemporaneous increases in the cost of debt financing (rather than strategic
considerations). For this to be the case, cartel formation would have to cause (or be caused by)
increases in the cost of debt financing. For example, banks may fear a reputation loss if a cartel is
detected and prosecuted, and they may also worry about a convicted cartel member’s ability to
service its debt. This assumes that a bank can detect a cartel while antitrust authorities cannot; and
this may further increase a bank’s required return if it could later be accused of being a co-
conspirator or facilitator of a cartel. On the other hand, if lenders are kept uninformed about the
cartel’s existence, any reputational and payment-risk considerations should arise only in post-
collusion years.
In order to capture the possible information effect during collusion years, we focus our
analysis on relationship lending, using data on private loan contracting terms from the Loan Pricing
Corporation’s (LPC) Dealscan database. The Dealscan database contains detailed loan
information for U.S. and foreign commercial loans made to government entities and
corporations.15 Merging the Dealscan data with our main database causes significant sample
attrition, since loan data is only available in years in which our sample firms signed new loan
with firm size deciles as instruments. The coefficient for Collusion is almost identical to that reported in Table 5, but
now it is statistically significant at the 5% level instead of the 10% level. 15 For a detailed description of this database see, for example, Chava and Roberts (2008).
21
contracts. We are left with close to 20,000 firm-year observations (570 of them correspond to
cartel firm-years).
We focus on two characteristics of loans that are associated with debt financing being more
“costly”: The coupon offered to the lender, and whether the loan is secured. Debt financing is less
costly to a borrower if the coupon rate is lower and if the loan is not secured (offering collateral
reduces a firm’s debt capacity). We define Spread as the “all-in-drawn” spread (in basis points)
over LIBOR, computed as the sum of coupon and annual fees on the loan in excess of six-month
LIBOR. The average Spread in our sample is 191 basis points. And we define Secured as a
dummy that takes a value of 1 if the loan is secured, and 0 otherwise; 73% of the loan-years in our
sample are secured.
If cartel formation leads to reduced leverage through a cost-of-debt channel, then cartel
formation must (1) be observable to a lender and (2) reduce a cartel member’s credit quality. The
cost of debt financing and the use of collateral should then increase during collusion years. In
contrast, the Commitment Hypothesis predicts that the cost of debt financing and the use of
collateral should not be affected during collusion years, or that they decrease, since cartel firms
reduce their leverage below the level they would otherwise find optimal. The predictions for the
post-collusion years are less clear. If there are reputation effects, then we would expect that post-
collusion cartel firms face a higher cost of debt financing and offer collateral more frequently.
However, they may choose to borrow less.
Table 6 presents the results. When estimating either Spread or Secured, the coefficient for
Collusion is insignificant. So there is no evidence either that lenders are informed about a
borrower’s membership in a cartel, or that this has an adverse effect on their cost of debt financing.
In other words, the evidence is inconsistent with the reduction in leverage during collusion
operating through a cost-of-debt channel. However, there are significant effects in post-collusion
years: The coefficients for Post Collusion are both significant and large. This suggests that
convictions for membership in a cartel have significant negative effects that operate through a cost-
of-debt channel. This finding may explain why in our leverage regressions above, the coefficient
22
for Post Collusion tended to be negative (cartel firms had lower leverage after cartels were
dissolved because of an increased cost of debt financing).16
Table 6
5. Collusion and Other Financial Policies
In this section, we analyze a broader set of financial policies. This allows us to shed light
on other dimensions of the strategic behavior of cartel firms, which in turn can further substantiate
our inferences about the Commitment Hypothesis. Specifically, we ask how cartel firms use their
increased profits, by studying the payout policies and cash holdings of cartel firms during collusion
and post-collusion years. We also discuss changes in capital expenditure and R&D, because they
represent important uses of cash. Furthermore, these additional results help us rule out possible
alternative explanations for our findings (which we address in Section 6) based on links between
profitability and leverage that are different from the Commitment Hypothesis or the Trade-off
Theory.
5.1. Collusion and Payout Policy
As described in Table 1, cartel firms have higher total payout ratios than other firms. It is
thus likely that the profit increases during collusion years are passed on to shareholders. It is
furthermore likely that the cartel firms would use share repurchases instead of dividend increases,
for several reasons. First, while cartel firms are more likely to pay dividends than other firms, the
average dividend paid is relatively low (see Table 1, Panel B). The higher payout is achieved by
having more active share repurchase programs than other firms. Second, the need to keep a cartel
agreement (and its effects) secret may let cartel firms favor share repurchases. Third, if a cartel
seems unstable and may break up in the near future, then cartel firms should regard the collusion
profits as temporary and thus favor repurchases over dividend increases.
We define Total Payout as the sum of the dividends paid and the amounts of common and
preferred stock repurchased, divided by the lagged value of total assets. We then estimate Equation
(1) with Total Payout as a dependent variable (instead of Leverage). Following prior work on the
16 In unreported analyses we study the maturity of the loans and find it is unrelated to collusion. The same holds for
the maturity structure of all debt, based on Compustat data.
23
determinants of payout policies (see, e.g., Fama and French (2001) or Hoberg et al. (2014)) we use
the following control variables: Lagged profitability, lagged sales, cash flow volatility, and the
market to book ratio (MB). Next, we distinguish dividends from repurchases. We define
Dividends and Repurchases as the cash dividends divided by the lagged value of total assets and
the amounts of common and preferred stock repurchased divided by the lagged value of total
assets, respectively.
Table 7 shows the results of estimating Equation (1) with Total Payout as a dependent
variable. Column (1) shows the results without using controls while Column (2) shows the results
using the controls listed above. The coefficients for Collusion are statistically and economically
significant: Cartel firms increase their total payout by 0.7%-0.8% of their assets, a large increase
considering that the average payout is 1.3% of total assets (see Table 1 above). Columns (3) and
(4) of Table 7 show the corresponding results distinguishing dividends from repurchases. The
increase in total payout is clearly driven by increases in share repurchases. The only effect
collusion has on dividends is that dividends decrease after a cartel is dissolved, which is not
surprising if this reintroduces competition to the industry and thus reduces profits.
Table 7
We also examine whether the competitive environment affects cartel firms’ payout during
collusion. Hoberg et al. (2014) show that the strength of the competitive pressure a firm faces
negatively affects how much of its profits it pays out. This is consistent with the findings in
Haushalter et al. (2007) that firms in more competitive environments hoard cash for precautionary
reasons (to protect themselves against predation, or to exploit new opportunities), which means
that the payout to shareholders is likely lower. If this intuition is valid, we expect that any increases
in payout during collusion periods should be more pronounced in less competitive environments,
where cartels are more stable and the risk of the cartel breaking up is likely lower.
We test this prediction by splitting the sample into subsamples with above-median and
below-median Product-Market Fluidity, as we did in Section 4.4.2 above. We present the results
in Table 8, columns (1) and (2). We find that firms facing weaker competitive pressure (low
fluidity) substantially increase their payout during collusion, whereas firms facing stronger
competitive pressure (higher fluidity) do not. That is consistent with our predictions: Firms in more
24
competitive environments face more competitive threats; their cartels are likely less stable, and if
collusion breaks down, they likely face stronger competition. Thus, they are less likely to
distribute cash to shareholders during collusion periods and instead retain cash for precautionary
reasons.
Table 8
Overall, the evidence suggests that cartels change their payout during collusion years, and
that they choose these changes strategically. This is consistent with the Commitment Hypothesis,
in the sense that cartel firms strategically change their overall financial strategies during collusion
periods.
5.2. Collusion and Cash Holdings
We now examine how collusion affects the cash holding decisions of cartel firms. As
discussed above, firms facing stronger competitive pressure should hold larger cash balances for
precautionary reasons (Haushalter et al. 2007). Collusion by design reduces or eliminates
competitive pressure, so there is less need to build up cash balances for precautionary reasons.
However, collusion leads to higher profits, which make it possible to quickly accumulate
significant cash balances. It is not clear a priori what effect collusion should have on cash balances.
We estimate Equation (1), with Cash Holdings as a dependent variable. We present the
results in Table 9. Column (1) presents the results without controls, while Column (2) present the
results controlling for variables commonly used in the cash holdings literature (see, e.g., Bates et
al. 2009). The results indicate that cash holdings decrease by 1-1.4% of firms’ assets, a significant
decrease given average cash holdings (for cartel firms) of 7% (see Table 1); but the coefficients
are either marginally significant (p-value of 9.7%) or statistically insignificant (p-value of 12.1%).
These results suggest that collusion may cause decreases in the cash holdings of cartel firms, but
the results are not as strong as our earlier results. Given the ambiguous predictions, that is not
surprising.
The accumulation of cash balances for precautionary reasons (Haushalter et al. 2007)
should be less beneficial for firms that face weaker competitive pressure. We thus split the sample
by Product-Market Fluidity, as before, and repeat the regressions for the two subsamples. The
25
results are in columns (3) and (4) of Table 9. There are no significant changes in cash holdings
during collusion years or afterwards for firms that face strong competitive pressure (high fluidity).
Importantly, the cash holdings of cartel firms facing weak competitive pressure (low fluidity)
decrease during collusion years. This is consistent with the empirical findings in Haushalter et al.
(2007) and with our payout results. These results further highlight that firms change their financial
policies for strategic reasons during collusion periods, and thus that our findings are unlikely to be
driven by other confounding factors.
Table 9
5.3. Collusion, Capital Expenditure and R&D Expense
We now study whether cartel firms change their investment decisions during collusion
periods. Specifically, we study changes in capital expenditure and R&D, as they are important uses
of cash and they can shed light on the inner workings of cartels. Table 10 presents the results.
Columns (1) and (2) show that during collusion periods, capital expenditure (normalized by assets)
does not change in a significant way. That is not surprising: The purpose of the cartels is to limit
supply and raise prices, so investments in expanding capacity would be counter-productive since
they threaten the stability of a cartel. (A firm with larger spare capacity would find it easier and
more attractive to undercut its fellow cartel members’ prices.)
Table 10
We find that R&D expenses (also normalized by assets) are higher during the collusion
periods, see columns (3) to (5) in Table 10. In the regression reported in columns (3) and (4), we
include all firm-years: We substitute missing R&D values by zero (as is common in the literature,
we interpret the absence of R&D expense in the data as indicating that R&D expense is zero or
negligible.). In the regression reported in column (5) we only use firms for which R&D data is
originally available in at least one year. That R&D is higher during collusion periods is consistent
with Dasgupta and Stiglitz (1980) and Phillips and Zhdanov (2013), who predict (and empirically
confirm) that firms facing reduced competition invest more in R&D, since it is more likely that
they can internalize the benefits from such investments in those environments.
26
6. Alternative Explanations
6.1. Long-run risk
One possible concern with our findings is that the formation of a cartel may somehow
increase the risk faced by a cartel firm. For example, one could argue that the threat of the cartel
failing and competition resuming influences the decisions firms make during collusion periods. If
cartel formation does indeed increase risk, then a natural response would be to reduce a cartel
firm’s financial leverage.
This alternative explanation, however, conflicts with several of our results. Firms do not
seem to prepare for bad outcomes. They do not increase cash holdings but instead decrease them
or leave them unchanged. Also inconsistent with the risk explanation is that cartel firms increase
their payout during collusion periods, and that during recessions the reduction in leverage is less
pronounced in collusion periods. Furthermore, we find that capital expenditure does not change,
and that R&D expenses increase, both inconsistent with firms being exposed to more risk. Finally,
our IV regressions, which are less affected by such potential unobserved confounding factors, also
show a reduction in leverage. Overall, there is no reason to believe that the threat of possible
resumption of competition affects financial leverage choices during collusion periods.
6.2. Strategic Debt
The presence of “strategic debt” can create a link between profitability and leverage,
possibly leading to an alternative explanation of our findings. Brander and Lewis (1986) show
that financial leverage can make firms aggressive in their product markets, due to a risk-shifting
effect. They show that for an individual firm it may be optimal to lever up strategically, expecting
other firms to cut back their output in response. The model then predicts that more profitable firms
have higher leverage. But that prediction is inconsistent with the evidence we find.
However, if all firms in an industry can take on “strategic debt”, the outcome may resemble
the “bad equilibrium” in a prisoners’ dilemma game, with all firms producing high outputs, and
with correspondingly lower prices and profits. If so, the model predicts that high-leverage firms
have lower profits. That would be consistent with some of our findings. However, given that the
predictions about the relation between leverage and profits are ambiguous (and thus consistent
27
with any empirical findings), the model in Brander and Lewis (1986) cannot help explain our
findings.
The results in Brander and Lewis (1986) are derived for a model in which revenues are
risky (their intuition is that risk shifting drives the effects). One might thus argue that the results
should be stronger for firms that make risky investments, for example R&D-intensive firms. Note
that cartel firms have a lower R&D intensity than other firms (see Table 1, Panel B), so it is unlikely
that risky investments are driving our results. To test this more thoroughly, we split our sample
according to each firm’s R&D intensity, and run separate regressions using leverage as dependent
variable. We present the results in Table 11. We find that the coefficient for Collusion is
insignificant for high-R&D cartel firms, while it is significant for low-R&D cartel firms. Thus,
the possibility of making risky investments is not the driver of our findings.17 That the evidence
does not support Brander and Lewis (1986) is perhaps unsurprising since subsequent work has
shown that the predictions of that model are not robust (see Showalter 1995 and Povel and Raith
2004) and there are doubts about the empirical relevance of risk shifting explanations (see
Hernández-Lagos et al., 2016).
Table 11
6.3. Pecking Order
The Pecking Order Theory (Myers 1984) suggests that when it comes to financing new
investment, firms prefer using retained cash to issuing more debt, and they prefer issuing more
debt to issuing more equity. If profits increase for exogenous reasons, one would expect firms to
reduce their leverage somewhat, leading to a negative relation between profitability and leverage.
Higher profits may also lead to increased investment if a budget constraint has been relaxed. This
may moderate or even reverse the reduction in leverage.
If such an effect is present, we are controlling for it in our regressions, since they include
lagged profits as controls. So it is unlikely that pecking-order effects are driving our results.
17 If we include only firms with R&D data available and then split the sample by R&D intensity, the coefficient for
Collusion is insignificant for both subsamples.
28
The key element in the Pecking Order Theory is asymmetric information: the less well
informed investors are about a firm’s prospects and investment opportunities, the more costly it is
to use sources of funds farther down the pecking order. Frank and Goyal (2003) and Leary and
Roberts (2010) show that information asymmetry proxies do a poor job at explaining debt and
equity issuance as predicted by the Pecking Order Theory.18 Nevertheless, we can ask what effect
asymmetric information has on the leverage choices of cartel firms. Arguably, asymmetric
information is more likely to be a concern for R&D-intensive firms. Thus, to the extent that the
Pecking Order Theory is somewhat related to our findings, we should find that the reduction in
leverage is stronger for R&D-intensive firms. However, we find the opposite (see Table 11), which
further suggests that the Pecking Order Theory is unrelated to our findings.
The Pecking Order Theory does not make very specific predictions. For example, it is
unclear whether high-leverage firms should decrease their leverage more strongly if profits
increase. It is plausible, however, that high-leverage firms are financially constrained and issue
significant amounts of debt to finance attractive investment opportunities. If more internal funds
become available, then the primary effect may be that capital expenditure increases. However, we
find that capital expenditure does not increase for highly levered firms (see Table A.2 in the
Appendix).
In sum, the predictions of the Pecking Order Theory cannot explain our findings, or the
predictions are ambiguous and thus are unrelated to our findings.
6.4. Agency Problems
Managerial agency problems can affect the relation between profitability and leverage.
Kovenock and Phillips (1995, 1997) argue that if agency problems cause overinvestment, and debt
mitigates those agency problems (Jensen 1986), then high-debt firms should be more efficient. In
other words, there may be a positive relation between profitability and financial leverage. That
cannot explain our findings, however, since we find a negative relation.
18 Transaction costs could also generate a pecking order. But proxies for transaction costs also align poorly with the
predictions of the pecking order theory. For additional evidence against the pecking order theory explaining the
negative association between profits and leverage see Fama and French (2005) or Rauh and Sufi (2010).
29
The free cash flow problem analyzed in Jensen (1986) is more likely to be a concern if
governance is weak, including pressure to make firms efficient coming from product market
competition. In other words, if firms face stronger competitive threats, the free cash flow problem
is not likely to be a concern. We find stronger results for firms with higher product market fluidity,
so (again) the free cash flow problem cannot explain our findings.
In unreported tests we used sample splits to analyze whether changes in standard measures
of corporate governance (the G-Index (Gompers et al. 2003); the percentage of independent
directors; and the presence of a staggered board) affect leverage during collusion years, or whether
there are changes in these variables during collusion years. We found no significant effects in any
of those regressions, which suggests that agency problems cannot explain our results.
6.5. Growth Opportunities
As modeled in Strebulaev (2007), a negative relation between profitability and leverage
may arise in a dynamic model: Positive profitability shocks increase the value of a firm (by
increasing the present value of its growth options) and, if the firm does not continuously rebalance
its capital structure, this reduces the firm’s leverage ratio. Note that this only affects the firm’s
market value leverage ratio, but not its book value leverage ratio. In our tests, we measure leverage
using book values, precisely to avoid mechanical effects caused by changes in expectations and
corresponding changes in valuations.
How growth opportunities affect book value leverage is analyzed in Barclay et al. (2006).
They show that investments in long-term growth options have a negative debt capacity, since the
threat of underinvestment caused by debt overhang increases the cost of using debt, and since it is
less likely that the free cash flow problem (Jensen 1986) causes efficiency losses (and so the
benefits of having debt are reduced). Consequently, firms with more growth options (including
R&D-intensive firms) should have lower book leverage.19 This could potentially explain our
findings if growth options increase during collusion. However, this scenario is unlikely since
leverage stays low after collusion ends (when growth options should dissipate) and since the
19 Furthermore, investments in R&D often require follow-up investments in future years, thus tying up future cash
flows and reducing a firm’s debt capacity. For a dynamic model of financing and investment, see Lambrecht and
Myers (2016).
30
leverage reduction is statistically significant only for firms with low R&D (see Table 11 above),
which are arguably the firms with fewer growth options available.
6.6. Dynamic Trade-off Theory
An alternative to the static Trade-off theory is the dynamic Trade-off Theory model
proposed by Hennessy and Whited (2005). They study a firm’s decisions over multiple periods:
given the realized cash flow, it chooses its investment, dividends, share repurchases, and changes
in debt outstanding. This model can generate a negative association between profitability and
leverage and thus could potentially explain the reduction in leverage we find during collusive
periods. However, in their model, leverage is path dependent and exhibits hysteresis, i.e., firms
with larger debt in place reduce their leverage by less if they experience a positive profit shock,
since they have a larger financing gap. This prediction conflicts with our findings in Table 3, where
we find that firms with high initial leverage display larger reductions in leverage during collusion
years. The Hennessy and Whited (2005) model also does not explain why firms reduce their
leverage by more during boom periods and by less during recessions (see Table 4), and why the
reduction in leverage is stronger for firms that face larger competitive threats (firms with higher
product market fluidity, see Table 3).
Frank and Goyal (2015) also consider a dynamic model of leverage choices, incorporating
the costs of changing debt and equity levels. They conclude that the relation between profitability
and leverage may be negative. They find that profit increases lead to both debt increases and equity
increases, and that leverage decreases because the equity increases are larger. However, this
cannot explain our findings: We show in Figure 2 (Panel B) that the debt levels of cartel firms
decrease when collusion starts, i.e., when their profits increase.
7. Conclusions
This paper shows that cartel firms change their financial policies, in particular their
leverage ratios, during periods of active collusion. Specifically, our tests show that cartel firms
decrease their financial leverage, consistent with the intuition gained from Maksimovic (1988) that
low leverage levels make it easier for colluding firms to add stability to their illegal agreements.
31
One important lesson from this finding is that it can help resolve a puzzle in the capital
structure literature. Past empirical studies have found a negative relation between profitability and
financial leverage, which is inconsistent with what one would expect based on intuition from the
Trade-off Theory. The negative relation we find for cartel firms can help resolve this puzzle,
because our tests show that there are good reasons to expect the relation to be negative for cartel
firms, and because these firms represent a significant fraction of the economy. Obviously, this
cannot entirely explain the negative relation that has puzzled the literature, but our findings
suggests that more research into other drivers of profitability is called for.
The second important contribution of our paper is that it provides evidence on the financial
policies of cartel firms, and how these policies are used strategically to make collusive agreements
more stable. A better understanding of cartels is helpful because cartel firms have a surprisingly
large economic footprint, and because changes in financial policies represent a new symptom of
collusion and cartel activity that may help antitrust authorities detect and convict colluding firms.
32
References
Aggrawal, Anup, and Mark Chen, 2011. “Boardroom brawls: An empirical analysis of disputes
involving directors.” Working paper.
Artiga, Tanja, Markus Schmid, and David Yermack, 2013. “Smokescreen: How managers behave
when they have something to hide.” Working paper, New York University.
Barclay, Michael, Clifford Smith, and Erwan Morellec, 2006. “On the Debt Capacity of Growth
Options.” Journal of Business 79, 37-59.
Bates, Thomas, Kathleen Kahle, and René Stulz, 2009. “Why do U.S. firms hold so much more
cash than they used to?” Journal of Finance 64, 1985-2021.
Bolton, Patrick, and David Scharfstein, 1990. “A theory of predation based on agency problems
in financial contracting.” American Economic Review 80, 93-106.
Bos, Iwan, and Joseph Harrington, 2010. “Endogenous cartel formation with heterogeneous
firms.” RAND Journal of Economics 41, 92-117.
Bradley, Michael, Gregg Jarrell, and E. Han Kim, 1984. “On the existence of an optimal capital
structure: Theory and evidence.” Journal of Finance 39, 857-877.
Brander, James, and Tracy Lewis, 1986. “Oligopoly and Financial Structure: The Limited
Liability Effect.” American Economic Review 76, 956-970.
Bryant, Peter, and Woodrow Eckard, 1991. “Price Fixing: The Probability of Getting Caught.”
Review of Economics and Statistics 73, 531-536.
Campello, Murillo, Daniel Ferrés, and Gaizka Ormazabal, 2015. “Whistleblowers on the Board?
The Role of Independent Directors in Cartel Prosecutions”. Working Paper.
Chava, Sudheer, and Michael Roberts, 2008. “How does financing impact investment? The role
of debt covenant violations”. Journal of Finance 63, 2085-2121.
33
Ciliberto, Federico, and Jonathan Williams, 2014. “Does multimarket contact facilitate tacit
collusion? Inference on conduct parameters in the airline industry.” RAND Journal of
Economics 45, 764-791.
Connor, John, 2011. “Cartel Detection and Duration Worldwide.” Competition Policy
International: Antitrust Chronicle 9, 2-10.
Connor, John, 2014. “The Private International Cartels (PIC) Data Set: Guide and Summary
Statistics, 1990−2013.” Working paper, available at http://ssrn.com/abstract=2478271
Connor, John, and Yuliya Bolotova, 2006. “Cartel overcharges: Survey and meta-analysis.”
International Journal of Industrial Organization 24, 1109-1137.
Dasgupta, Partha, and Joseph Stiglitz, 1980. “Industrial structure and the nature of innovative
activity.” Economic Journal 90, 266–93.
Dasgupta, Sudipto, and Alminas Žaldokas, 2016. “Cartels Are Gone And So Is Leverage”.
Working paper.
Dong, Ailin, Massimo Massa, and Alminas Žaldokas, 2014. “Busted! Now What? Effects of Cartel
Enforcement on Firm Value and Corporate Policies”. Mimeo.
Fama, Eugene, and Kenneth French, 2001. “Disappearing dividends: Changing firm characteristics
or lower propensity to pay?” Journal of Financial Economics 60, 3-44.
Fama, Eugene, and Kenneth French, 2005. “Financing Decisions: Who issues stock?” Journal of
Financial Economics 76, 549-82.
Frank, Murray, and Vidhan Goyal, 2003. “Testing the pecking order theory of capital structure.”
Journal of Financial Economics 67, 218-48.
Frank, Murray, and Vidhan Goyal, 2015. “The profits-leverage puzzle revisited.” Review of
Finance 19, 1415-1453.
Friedman, James, 1971. “A noncooperative equilibrium for supergames.” Review of Economic
Studies 38, 1-12.
34
Gompers, Paul, Joy Ishii, and Andrew Metrick, 2003. “Corporate Governance and Equity Prices.”
Quarterly Journal of Economics 118, 1007-155.
Graham, John, and Mark Leary, 2011. “A Review of Empirical Capital Structure Research and
Directions for the Future.” Annual Review of Financial Economics 3, 309-345.
Grullon, Gustavo, and Roni Michaely, 2007. “Corporate payout policy and product market
competition”. Working paper.
Harrington, Joseph, 2006. “How Do Cartels Operate?” Foundations and Trends in
Microeconomics 2, 1-105
Haushalter, David, Sandy Klasa, and William Maxwell, 2007. “The influence of product market
dynamics on a firm’s cash holdings and hedging behavior.” Journal of Financial
Economics 84, 797-825.
Hennessy, Christopher, and Toni Whited, 2005. “Debt dynamics.” Journal of Finance 60, 1129-
65.
Hernández-Lagos, Pablo, Giorgo Sertsios, and Paul Povel, 2016. “An Experimental Analysis of
Risk Shifting Behavior.” Review of Corporate Finance Studies, forthcoming.
Hoberg, Gerard, and Gordon Phillips, 2010. “Product market synergies in mergers and
acquisitions: A text based analysis.” Review of Financial Studies 23, 3773–3811.
Hoberg, Gerard, and Gordon Phillips, 2016. “Text-Based Network Industries and Endogenous
Product Differentiation.” Journal of Political Economy, forthcoming,
Hoberg, Gerard, Gordon Phillips, and Nagpurnanand Prabhala, 2014. “Product Market Threats,
Payouts, and Financial Flexibility.” Journal of Finance 69, 293-324.
Jensen, Michael, 1986. “Agency cost of free cash flow, corporate finance, and takeovers.”
American Economic Review 76, 323-329.
Kovenock, Dan, and Gordon Phillips, 1995. “Capital structure and product-market rivalry: How
do we reconcile theory and evidence?” American Economic Review 85, 403-408.
35
Kovenock, Dan, and Gordon M. Phillips, 1997. “Capital structure and product market behavior:
An examination of plant exit and investment decisions.” Review of Financial Studies 10,
767-803.
Kraus, Alan, and Robert Litzenberger, 1973. “A State-Preference Model of Optimal Financial
Leverage.” Journal of Finance 28, 911-922.
Lambrecht, Bart, and Stewart Myers, 2016. “The Dynamics of Investment, Payout and Debt.”
Working paper, Cambridge University and MIT.
Leary, Mark, and Michael Roberts, 2014. “Do peer firms affect corporate financial policy?”
Journal of Finance 69, 139-178.
Lemmon, Michael, Michael Roberts, and Jaime Zender, 2008. “Back to the beginning: Persistence
and the cross-section of corporate capital structure.” Journal of Finance 63, 1575-1608.
Levenstein, Margaret, and Valerie Suslow, 2006. “What Determines Cartel Success?” Journal of
Economic Literature 44, 43-95.
Levenstein, Margaret, and Valerie Suslow, 2011. “Breaking Up Is Hard to Do: Determinants of
Cartel Duration.” Journal of Law and Economics 54, 455-492.
Lewellen, Katharina, 2006. “Financing decisions when managers are risk-averse.” Journal of
Financial Economics 82, 551-589.
Lex Mundi, 2013. “Leniency Program Global Practice Guide.” Report prepared by the Antitrust,
Competition & Trade Practice Group, Lex Mundi, Houston. Available at
http://www.lexmundi.com/Document.asp?DocID=6464
Low, Angie, 2009. “Managerial risk-taking behavior and equity-based compensation.” Journal of
Financial Economics 92, 470-490.
MacKay, Peter, and Gordon Phillips, 2005. “How Does Industry Affect Firm Financial Structure?”
Review of Financial Studies 18, 1433-1466.
36
Maksimovic, Vojislav, 1988. “Capital Structure in Repeated Oligopolies.” RAND Journal of
Economics 19, 389-407.
Miller, Nathan, 2009. “Strategic Leniency and Cartel Enforcement.” American Economic Review
99, 750-768.
Myers, Stewart, 1984, “The Capital Structure Puzzle,” Journal of Finance 39, 575–592.
Myers, Stewart, 1993, “Still Searching for Optimal Capital Structure.” Journal of Applied
Corporate Finance 6, 4-14.
OECD, 2002. “Fighting Hard Core Cartels: Harm, Effective Sanctions and Leniency
Programmes.” Report, Organisation For Economic Co-Operation And Development, Paris.
Available at http://www.oecd.org/competition/cartels/1841891.pdf
Parsons, Christopher, and Sheridan Titman, 2008a. “Empirical Capital Structure: A Review.”
Foundations and Trends in Finance 3, 1-93.
Parsons, Christopher, and Sheridan Titman, 2008b. “Capital Structure and Corporate Strategy.”
Handbook of Corporate Finance, Empirical Corporate Finance. Ed. Espen Eckbo.
Elsevier North-Holland, Amsterdam.
Petersen, Mitchell, 2009. “Estimating Standard Errors in Finance Panel Data Sets: Comparing
Approaches.” Review of Financial Studies 22, 435-480.
Phillips, Gordon, 1995. “Increased debt and industry product markets: An empirical analysis.”
Journal of Financial Economics 37, 189-238.
Phillips, Gordon, and Alexei Zhdanov, 2013. “R&D and the Incentives from Merger and
Acquisition Activity.” Review of Financial Studies 26, 34-78.
Povel, Paul, and Michael Raith, 2004. “Financial Constraints and Product Market Competition:
Ex-ante vs. Ex-post Incentives.” International Journal of Industrial Organization 22, 917-
949.
37
Rauh, Joshua, and Amir Sufi, 2010. “Capital Structure and Debt Structure.” Review of Financial
Studies 23, 4242-4280.
Riordan, Michael, 2003. “How Do Capital Markets Influence Product Market Competition?”
Review of Industrial Organization 23, 179-191.
Rotemberg, Julio, and Garth Saloner, 1986. “A supergame-theoretic model of business cycles and
price wars during booms.” American Economic Review 76, 390-407.
Showalter, Dean, 1995. “Oligopoly and financial structure: Comment.” American Economic
Review 85, 647-653.
Symeonidis, George, 2003. “In Which Industries Is Collusion More Likely? Evidence from the
UK.” Journal of Industrial Economics 51, 45-74.
Xu, Jin, 2012. “Profitability and capital structure: Evidence from import penetration.” Journal of
Financial Economics 106, 427–446.
38
Figure 1
Panels A and B plot the coefficients, and 95% confidence intervals, of the following regression:
𝑦𝑖𝑡 = 𝛼 + 𝛽 ∗ 𝐶𝑜𝑙𝑙𝑢𝑠𝑖𝑜𝑛𝑖𝑡 + ∑γℎ ∗ 𝑑𝑖ℎ + 𝜑𝑖 + µ𝑡 + 𝜀𝑖𝑡 ,
The subscript h indexes the years that immediately precede collusion years (h {−3, −2, −1}), years that immediately
follow collusion years (h {1, 2, 3}), or years that are collusion years (h = 0 for full collusion years, h = 0– for a
partial collusion year at the start of a cartel, and h = 0+ for a partial collusion year at the end of a cartel). The indicator
variable dih takes a value of 1 if a firm operates in one of those years, and 0 otherwise. Panel A plots regression
coefficients h using book leverage as the dependent variable, while Panel B plots regression coefficients h using
ROA as the dependent variable.
Panel A: Coefficients 𝛄𝒉 𝐰𝐡𝐞𝐧 𝒚 = 𝑳𝒆𝒗𝒆𝒓𝒂𝒈𝒆
Panel B: Coefficients 𝛄𝒉 𝐰𝐡𝐞𝐧 𝒚 = 𝑹𝑶𝑨
-.0
4-.
02
0
.02
.04
-3 -2 -1 0- Full col. yr 0+ +1 +2 +3
Coefficients Leverage Regression
-.0
2-.
01
0
.01
.02
.03
-3 -2 -1 0- Full col. yr 0+ +1 +2 +3
Coefficients ROA Regression
39
Figure 2
Panel A of this figure plots the average leverage ratios of cartel firms, separately for each year immediately before the
start of a collusion period, and for each of the first few years of a collusion period. There is an apparent drop in
leverage during the first two years of a colluding period. Panel B plots the numerator and denonimator of the leverage
ratio, debt and total assets, for the same years, showing that while both decrease with the duration of collusion, the
initial drop in leverage is driven by a more significant drop in average debt.
Panel A: Leverage Around Initial Collusion Years
Panel B: Debt and Total Assets Around Initial Collusion Years
.24
.25
.26
.27
.28
Mea
n L
eve
rage
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to collusion
Leverage around initial collusion years
50
00
60
00
70
00
80
00
Mea
n T
ota
l Ass
et (m
illio
n)
14
00
15
00
16
00
17
00
18
00
19
00
Mea
n T
ota
l De
bt (m
illio
n)
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to collusion
Mean Total Debt (million) Mean Total Asset (million)
Debt and assets around initial collusion years
40
Table 1. Descriptive Statistics
This table reports descriptive statistics for our sample of firms prosecuted for cartel participation between 1985 and
2012, and for the control firms. Panel A presents detailed descriptive statistics for all cartel and control firms (pooled).
Panel B reports differences in selected characteristics of the cartel and the control firms. See Appendix A for variable
definitions. Differences significant at: *10%, **5% and ***1%.
Panel A: Summary Statistics
Variable Mean p10 p90 sd N
Leverage 0.33 0.00 0.66 0.47 129,253
Total Payout 0.013 0.000 0.030 0.04 123,191
Dividend 0.00049 0.0000 0.0002 0.00 123,191
Dividend Payer 0.39 0.00 1.0000 0.49 129,253
Equity Repurchases 0.01 0.00 0.03 0.04 123,191
Profitability 0.03 -0.29 0.22 0.23 129,253
Negative Prof 0.18 0.00 1.00 0.39 174,789
Tangibility 0.29 0.03 0.72 0.25 129,253
Sales (million) 1,017 2 2,364 2,818 129,253
Cash Flow Volatility 0.081 0.007 0.18 0.14 124,082
Cash 0.12 0.00 0.34 0.17 122,854
Assets (million) 1,432 3 3,052 4,391 129,253
MB 4.16 0.84 4.38 40.00 107,622
R&D 0.12 0.00 0.25 0.51 67,962
Net Working Capital -0.09 -0.14 0.59 7.68 125,085
Panel B: Univariate Analysis, Cartel vs. Non-cartel Firms
Variable Cartel Non-Cartel Difference
Leverage 0.27 0.33 -0.06***
Total Payout 0.023 0.012 0.012***
Dividend 0.00002 0.0005 -0.0004***
Dividend Payer 0.73 0.38 0,35***
Equity Repurchases 0.02 0.01 0.01***
Profitability 0.14 0.02 0.12***
Negative Prof 0.03 0.25 -0,22***
Tangibility 0.34 0.29 0.05***
Sales (million) 6,220 959 5,261***
Cash Flow Volatility 0.026 0.082 -0.056***
Cash 0.07 0.13 -0.06***
Assets (million) 7,688 1,362 6,325***
MB 2.02 4.19 -2,16**
R&D 0.03 0.12 -0,09***
Net Working Capital 0.16 -0.09 0.25
41
Table 2. Collusion and Capital Structure
This table presents the results of analyzing the association between collusion and leverage. The dependent variable
Leverage is total liabilities divided by book value of assets. Panel A presents results using cartel firms: Columns (1)
and (2) present the estimation results using both cartel and control firms. Columns (3) and (4) present the estimation
results using only cartel firms. Panel B presents the results using only firms from industries in which there was
collusion, but the firms were not cartel members (i.e., they were the cartel firms’ non-colluding rivals). Collusion
takes a value of 1 for cartel firms during collusion years, and 0 otherwise; Post Collusion takes a value of 1 for cartel
firms during the 5 years after a cartel is dissolved, and 0 otherwise. The Controls include: Profitability; Tangibility;
Cash Flow Volatility; and Sales (see Appendix A for definitions). Standard errors (in parentheses) are adjusted for
heteroscedasticity and clustered at the industry level. Significant at: *10%, **5% and ***1%.
Panel A: Cartel Firms and Control Firms
Dependent variable: Leverage
Cartel and control firms Cartel firms only
Independent Variables: (1) (2) (3) (4)
Collusion -0.0277** -0.0281** -0.0266** -0.0280**
(0.0115) (0.0111) (0.0131) (0.0127)
Post Collusion -0.0256 -0.0264 -0.0289 -0.0348
(0.0201) (0.0199) (0.0238) (0.0234)
Controls No Yes No Yes
Firm Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
SIC Cluster Yes Yes Yes Yes
R-squared (within) 0.0036 0.0281 0.0749 0.1057
N 129,253 129,252 1,427 1,427
Panel B: Non-Cartel Firms from Industries with Cartels
Dependent variable: Leverage
Independent Variables: (1) (2)
Collusion 0.0109 0.0146**
(0.0086) (0.0068)
Post Collusion 0.0208* 0.0210***
(0.0104) (0.0075)
Controls No Yes
Firm Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
SIC Cluster Yes Yes
R-squared (within) 0.0201 0.0599
N 16,390 15,467
42
Table 3. Collusion and Capital Structure: Sample Splits
This table presents results of analyzing the association between collusion and leverage, using sample splits. The
dependent variable Leverage is total liabilities divided by book value of assets. Product-Market Fluidity is a text-
based measure of product market competitiveness formulated by Hoberg and Phillips (2010) that uses product
descriptions in SEC Form 10-K filings. Columns (1) and (2) present the results of splitting the sample into firms with
below-median and above-median Leverage as measured in the first year of each firm’s observations in the data.
Columns (3) and (4) present the results of splitting the sample into firms with below-median and above-median
Product-Market Fluidity. Collusion takes a value of 1 for cartel firms during collusion years, and 0 otherwise; Post
Collusion takes a value of 1 for cartel firms during the 5 years after a cartel is dissolved, and 0 otherwise. The Controls
include: Profitability; Tangibility; Cash Flow Volatility; and Sales (see Appendix A for definitions). Standard errors
(in parentheses) are adjusted for heteroscedasticity and clustered at the industry level. Significant at: *10%, **5%
and ***1%.
Dependent variable: Leverage
Split by Initial Leverage Split by Product-Market Fluidity
Independent Variables:
Below Median
(1)
Above Median
(2)
Below Median
(3)
Above Median
(4)
Collusion 0.0093 -0.0603*** -0.0051 -0.0433**
(0.0134) (0.0176) (0.0132) (0.0199)
Post Collusion 0.0306 -0.0773** 0.0277 -0.0796**
(0.0190) (0.0380) (0.0216) (0.0321)
Controls Yes Yes Yes Yes
Firm Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
SIC Cluster Yes Yes Yes Yes
R-squared (within) 0.0390 0.0422 0.0347 0.0345
N 64,421 64,831 50,004 35,623
43
Table 4. Collusion and Capital Structure: Triple Differences
This table presents results of analyzing the association between collusion and leverage, using a triple-differences
approach. The dependent variable Leverage is total liabilities divided by book value of assets. Recession Year is a
dummy variable that takes a value of 1 in the recession years, as defined using the NBER recession year list. High
Leniency is a dummy variable that takes a value of 1 if in a given year a firm is operating a cartel that covers three or
more countries with leniency laws. Column (1) presents the estimation results for the triple-difference analysis that
exploits the variation in recession versus non-recession years. Column (2) presents the estimation results for the triple-
difference analysis that exploits the variation in a firm’s exposure to leniency laws. Collusion takes a value of 1 for
cartel firms during collusion years, and 0 otherwise; Post Collusion takes a value of 1 for cartel firms during the 5
years after a cartel is dissolved, and 0 otherwise. The Controls include: Profitability; Tangibility; Cash Flow Volatility;
and Sales (see Appendix A for definitions). We do not report the coefficient for Recession, since that dummy variable
is collinear with the year fixed effects. Standard errors (in parentheses) are adjusted for heteroscedasticity and
clustered at the industry level. Significant at: *10%, **5% and ***1%.
Dependent variable: Leverage
Independent Variables:
Recession vs
Non-recession
(1)
High vs low exposure
to leniency laws
(2)
Collusion -0.0383*** -0.0282**
(0.0106) (0.0130)
Post Collusion -0.0245 -0.0322
(0.0202) (0.0238)
Recession Year*Collusion 0.0403***
(0.0112)
Recession Year*Post Collusion -0.0062
(0.0119)
High Leniency 0.0401
(0.0357)
High Leniency*Collusion -0.0568**
(0.0261)
High Leniency*Post Collusion -0.0807*
(0.0442)
Controls Yes Yes
Firm Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
SIC Cluster Yes Yes
R-squared (within) 0.0281 0.1103
N 129,252 1,427
44
Table 5. Collusion and Capital Structure: Instrumental Variables
This table presents results of analyzing the impact of collusion on leverage, using an instrumental variables approach. In the first stage, the dependent variables are
Collusion (takes a value of 1 for cartel firms during the cartel years, and 0 otherwise) and Post Collusion (takes a value of 1 for cartel firms during the 5 years after
a cartel is dissolved, and 0 otherwise). The instrumental variables for Collusion and Post Collusion are Cartels Active (using the number of firms colluding in years
t, t-1 and t-2, in the same 2-digit SIC code, but different 4-digit SIC code as a given firm) and Cartels Dissolved (using the number of firms ending a collusive
agreement in years t, t-1 and t-2, in the same 2-digit SIC code, but different 4-digit SIC code as a given firm). Columns (1) and (2) present the relation between the
instruments and Collusion and Post Collusion. Additionally, we classify cartel firms according to their assets in four quartile sizes, and interact quartile dummies
with the instruments. Columns (3) and (4) show the results of this alternative first-stage specification. In the second stage, the dependent variable Leverage is total
liabilities divided by book value of assets. Column (5) presents the results of the second-stage estimation, using the specification in Columns (3) and (4) as the
first-stage regressions. The Controls include: Profitability; Tangibility; Cash Flow Volatility; and Sales (see Appendix A for definitions). Standard errors (in
parentheses) are adjusted for heteroscedasticity and clustered at the industry level. Significant at: *10%, **5% and ***1%.
First Stage Second Stage
Instruments:
# Ongoing and Ended Cartels in Related
Industries
Instruments:
# Ongoing and Ended Cartels in
Related Industries Interacted with
Firms' Size Quartiles
Using
all
Instruments
Independent Variables:
Dep. Var:
Collusion
(1)
Dep. Var:
Post Collusion
(2)
Dep. Var:
Collusion
(3)
Dep. Var:
Post Collusion
(4)
Dep. Var:
Leverage
(5)
Collusion -0.3815*
(0.2143)
Post Collusion -0.2693
(0.1986)
Cartels Active 0.0149*** -0.0116** 0.0016* 0.0004
(0.0053) (0.0045) (0.0009) (0.0011)
Cartels Active * Size_q2 0.0027** -0.0019**
(0.0011) (0.0007)
Cartels Active * Size_q3 0.0113** -0.0120***
(0.0050) (0.0044)
Cartels Active * Size_q4 0.0479*** -0.0415***
(0.0145) (0.0115)
45
Cartels Dissolved -0.0071** 0.0140*** 0.0000 -0.0020**
(0.0031) (0.0049) (0.0005) (0.0008)
Cartels Dissolved * Size_q2 -0.0027** 0.0024**
(0.0013) (0.0012)
Cartels Dissolved * Size_q3 -0.0059 0.0112***
(0.0040) (0.0041)
Cartels Dissolved * Size_q4 -0.0290** 0.0589***
(0.0136) (0.0174)
Controls Yes Yes Yes Yes Yes
Firm Fixed Effects Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes
SIC Cluster Yes Yes Yes Yes Yes
F-Test (weak instruments) 139 141
N 115,367 115,367 115,367
46
Table 6. Collusion and Loan Contracting
This table presents results of analyzing the association between collusion and the terms of loan contracting. In Column
(1) the dependent variable Spread is the “All-in-Drawn" spread (in basis-points) over LIBOR, computed as the sum
of coupon and annual fees on the loan in excess of six-month LIBOR. In Column (2), the dependent variable Secured
is a dummy variable that takes a value of 1 if the (new or renewed) loan is secured, and zero otherwise. Collusion
takes a value of 1 for cartel firms during collusion years, and 0 otherwise; Post Collusion takes a value of 1 for cartel
firms during the 5 years after a cartel is dissolved, and 0 otherwise. The Controls include: Profitability; Tangibility;
Cash Flow Volatility; and Sales (see Appendix A for definitions). Standard errors (in parentheses) are adjusted for
heteroscedasticity and clustered at the industry level. Significant at: *10%, **5% and ***1%.
Independent Variables:
Dep. Var:
Spread
(1)
Dep. Var:
Secured
(2)
Collusion -1.7482 -0.0067
(9.5455) (0.0539)
Post Collusion 27.6938** 0.1448**
(12.2625) (0.0650)
Controls Yes Yes
Firm Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
SIC Cluster Yes Yes
R-squared (within) 0.2453 0.0394
N 19,566 14,624
47
Table 7. Collusion and Payout Policy
This table presents results of analyzing the association between collusion and payout policy. In Panel A, the dependent
variable Total Payout is the sum of dividends paid and the amounts of common and preferred stock repurchased,
divided by the lagged value of total assets. Columns (1) and (2) present the estimation results of the tests including all
cartel and control firms. Columns (3) and (4) present the estimation results of the tests including only the cartel firms.
Collusion takes a value of 1 for cartel firms during collusion years, and 0 otherwise; Post Collusion takes a value of 1
for cartel firms during the 5 years after a cartel is dissolved, and 0 otherwise. The Controls include: Profitability; Cash
Flow Volatility; MB and Sales (see Appendix A for definitions). Standard errors (in parentheses) are adjusted for
heteroscedasticity and clustered at the industry level. Significant at: *10%, **5% and ***1%.
Breakdown of Total Payout
Dep. Var:
Total Payout
Dep. Var:
Dividends
Dep. Var:
Repurchases
Independent Variables: (1) (2) (3) (4)
Collusion 0.0078*** 0.0072** -0.0000 0.0072**
(0.0028) (0.0028) (0.0000) (0.0028)
Post Collusion 0.0024 0.0029 -0.0001* 0.0030
(0.0032) (0.0034) (0.0000) (0.0034)
Controls No Yes Yes Yes
Firm Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
SIC Cluster Yes Yes Yes Yes
R-squared (within) 0.0105 0.0147 0.0125 0.0150
N 123,191 105,861 105,861 105,861
48
Table 8. Collusion and Payout Policy: Sample Split
This table presents results of analyzing the association between collusion and corporate payout policies, using a sample
split. The dependent variable Total Payout is the sum of dividends paid and the amounts of common and preferred
stock repurchased, divided by the lagged value of total assets. Product-Market Fluidity is a text-based measure of
product market competitiveness formulated by Hoberg and Phillips (2010) that uses product descriptions in SEC Form
10-K filings. Columns (1) and (2) present the results of splitting the sample into firms with below-median and above-
median Product-Market Fluidity. Collusion takes a value of 1 for cartel firms during collusion years, and 0 otherwise;
Post Collusion takes a value of 1 for cartel firms during the 5 years after a cartel is dissolved, and 0 otherwise. The
Controls include: Profitability; Cash Flow Volatility; MB and Sales (see Appendix A for definitions). Standard errors
(in parentheses) are adjusted for heteroscedasticity and clustered at the industry level. Significant at: *10%, **5%
and ***1%.
Dependent Variable: Total Payout
Split by Product-Market Fluidity
Below Median Above Median
Independent Variables: (1) (2)
Collusion 0.0084** 0.0031
(0.0034) (0.0044)
Post Collusion 0.0026 0.0034
(0.0039) (0.0063)
Controls Yes Yes
Firm Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
SIC Cluster Yes Yes
R-squared (within) 0.0259 0.0166
N 46,210 31,199
49
Table 9. Collusion and Cash Holdings
This table presents results of analyzing the impact of collusion on cash holdings. The dependent variable Cash is cash
and equivalents divided by total assets. Product-Market Fluidity is a text-based measure of product market
competitiveness formulated by Hoberg and Phillips (2010) that uses product descriptions in SEC Form 10-K filings.
Columns (1) and (2) present the estimation results using both cartel and control firms. Columns (3) and (4) present
the results of splitting the sample into firms with below-median and above-median Product-Market Fluidity. Collusion
takes a value of 1 for cartel firms during collusion years, and 0 otherwise; Post Collusion takes a value of 1 for cartel
firms during the 5 years after a cartel is dissolved, and 0 otherwise. The Controls include: MB; RD; Net Working
Capital; Negative Prof; Dividend Payer, Cash Flow Volatility; Sales (see Appendix A for definitions). Standard errors
(in parentheses) are adjusted for heteroscedasticity and clustered at the industry level. Significant at: *10%, **5%
and ***1%.
Dependent Variable: Cash
Split by Product-Market Fluidity
Pooled observations Below Median Above Median
Independent Variables: (1) (2) (3) (4)
Collusion -0.0098* -0.0137 -0.0173* 0.0060
(0.0058) (0.0088) (0.0096) (0.0205)
Post Collusion 0.0049 0.0031 0.0007 0.0022
(0.0088) (0.0093) (0.0113) (0.0155)
Controls No Yes Yes Yes
Firm Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
SIC Cluster Yes Yes Yes Yes
R-squared (within) 0.0089 0.0604 0.0792 0.0784
N 122,854 56,293 24,740 18,561
50
Table 10. Collusion and Investment
This table presents the results of analyzing the association between collusion and investment, either in R&D or capital
expenditure. It shows the results of regressing investment (capital expenditure or R&D) on various right-hand
variables. The dependent variable in columns (1) and (2) is capital expenditure divided by book value of assets; the
dependent variable in columns (3) to (5) is R&D divided by book value of assets. Column (5) uses only firms for
which R&D data is available in at least one year. Collusion takes a value of 1 for cartel firms during collusion years,
and 0 otherwise; Post Collusion takes a value of 1 for cartel firms during the 5 years after a cartel is dissolved, and 0
otherwise. The Controls include: Profitability; Tangibility; Cash Flow Volatility; and Sales (see Appendix A for
definitions). Standard errors (in parentheses) are adjusted for heteroscedasticity and clustered at the industry level.
Significant at: *10%, **5% and ***1%.
Dependent Variable: Capex
Dependent Variable: R&D
Dep. Var: R&D,
if positive
Independent Variables: (1) (2) (3) (4) (5)
Collusion -0.0047 0.0034 0.0028** 0.0028** 0.0054**
(0.0152) (0.0143) (0.0013) (0.0013) (0.0022)
Post Collusion 0.0129 0.0036 0.0007 0.0007 0.0030
(0.0222) (0.0205) (0.0021) (0.0021) (0.0032)
Controls No Yes No Yes Yes
Firm Fixed Effects Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes
SIC Cluster Yes Yes Yes Yes Yes
R-squared (within) 0.0463 0.1688 0.0052 0.0353 0.0530
N 115,387 115,386 129,253 129,252 63262
51
Table 11. Collusion and R&D Intensity
This table presents the results of analyzing the association between collusion and R&D intensity. It shows the results
of a split-sample regression, distinguishing firms for which R&D data is not available (column (1)) from firms for
which R&D data is available in at least one year (column (2)). Collusion takes a value of 1 for cartel firms during
collusion years, and 0 otherwise; Post Collusion takes a value of 1 for cartel firms during the 5 years after a cartel is
dissolved, and 0 otherwise. The Controls include: Profitability; Tangibility; Cash Flow Volatility; and Sales (see
Appendix A for definitions). Standard errors (in parentheses) are adjusted for heteroscedasticity and clustered at the
industry level. Significant at: *10%, **5% and ***1%.
Dependent variable: Leverage
Split by R&D Expense
Independent Variables:
No R&D
(1)
Positive R&D
(2)
Collusion -0.0480*** -0.0148
(0.0160) (0.0138)
Post Collusion -0.0478 -0.0112
(0.0303) (0.0231)
Controls Yes Yes
Firm Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
SIC Cluster Yes Yes
R-squared (within) 0.0258 0.0320
N 65,990 63,262
52
Appendix
Variable Definitions
Firm Characteristics:
Leverage Total liabilities divided by book value of assets
Total Payout Dividends paid plus common and preferred stock purchased divided by lagged book
value of assets
Dividends Cash dividends divided by lagged book value of assets
Dividend Payer Indicator variable that takes a value of one if Dividends is greater than zero, and zero
otherwise
Repurchases Value of common and preferred stock purchased divided by lagged book value of assets
Profitability Operating income before depreciation divided by book value of assets (ROA)
Negative Prof Indicator variable that takes a value of one if Profitability is less than zero, and zero
otherwise
Tangibility Net property, plant, and equipment divided by book value of assets
Sales Value of total sales measured in millions of U.S. dollars
Cash Flow Volatility Standard deviation of Profitability over the prior 3-year period
Cash Cash and equivalents divided by book value of assets
Assets Book value of assets measured in millions of U.S. dollars
MB Sum of the market value of equity and total liabilities divided by book value of assets
RD Research and development expenses divided by book value of assets
Net Working Capital Difference between current assets and current liabilities divided by book value of assets
Loan Characteristics:
Spread “All-in-Drawn" spread (in basis-points) over LIBOR, computed as the sum of coupon
and annual fees on the loan in excess of six-month LIBOR.
Secured Indicator variable that takes a value of 1 if the firm took out a secured loan in a year,
and 0 otherwise
Sample Splits:
Collusion Indicator variable that takes the value of 1 for cartel firms during the cartel years, and 0
otherwise
Post Collusion Indicator variable that takes a value of 1 for cartel firms during the 5 years after a cartel
is dissolved, and 0 otherwise
Product-Market
Fluidity
Text-based measure of product market competitive pressure formulated by Hoberg and
Phillips (2010), using product descriptions in 10-K filings
Recession Year Indicator variable that takes a value of 1 in a recession year, and 0 otherwise, classified
using the NBER recession year list
High Leniency Indicator variable that takes a value of 1 if in a given year a firm is operating in cartels
in three or more countries with leniency laws in place, and 0 otherwise
53
Instrumental Variables:
Cartels Active Natural logarithm of one plus the number of firms colluding in years t, t-1 and t-2, in
the same 2-digit SIC code, but different 4-digit SIC code as a given firm
Cartels Dissolved Natural logarithm of one plus the number of firms ending a collusive agreement in years
t, t-1 and t-2, in the same 2-digit SIC code, but different 4-digit SIC code as a given firm
54
Figure A.1
This figure plots the average of the correlation coefficient of the profitability of all members of a cartel, averaged over
all cartels, distinguishing firm-years before collusion, during collusion, and after collusion.
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
Pre-Collusion Collusion Post-Collusion
ROA correlation between colluding firms
55
Figure A.2
This figure plots the coefficients, and 95% confidence intervals, of the following regression:
𝑦𝑖𝑡 = 𝛼 + 𝛽 ∗ 𝐶𝑜𝑙𝑙𝑢𝑠𝑖𝑜𝑛𝑖𝑡 + ∑γℎ ∗ 𝑑𝑖ℎ + 𝜑𝑖 + µ𝑡 + 𝜀𝑖𝑡 ,
The subscript h indexes the years that immediately precede collusion years (h {−3, −2, −1}), years that immediately
follow collusion years (h {1, 2, 3}), or years that are collusion years (h {1,2,3,other} for full collusion years, h =
0– for a partial collusion year at the start of a cartel, and h = 0+ for a partial collusion year at the end of a cartel). The
indicator variable dih takes a value of 1 if a firm operates in one of those years, and 0 otherwise. The figure plots
regression coefficients h using book leverage as the dependent variable. The difference, compared with Figure 1,
Panel A, is that here we decompose the full-year collusion years in the first full three colluding years, and the other
colluding years.
-.0
6-.
04
-.0
2
0
.02
.04
-3 -2 -1 0- col1 col2 col3 oth. col 0+ +1 +2 +3
Coefficients Leverage Regression
56
Table A.1. Instrumental Variables, Decile Interactions
This table presents results of analyzing the impact of collusion on leverage, using an instrumental variables approach.
The first stage is estimated as in Table 5, columns (3) and (4), except that the variables Cartels Active and Cartels
Dissolved are interacted with size decile dummies instead of size quartile dummies. The results are not reported due
to space limitations. In the second stage, the dependent variable Leverage is total liabilities divided by book value of
assets. The Controls include: Profitability; Tangibility; Cash Flow Volatility; and Sales (see Appendix A for
definitions). Standard errors (in parentheses) are adjusted for heteroscedasticity and clustered at the industry level.
Significant at: *10%, **5% and ***1%.
Second Stage
Using all Instruments
Independent Variables: Dependent Variable: Leverage
Collusion -0.3838**
(0.1929)
Post Collusion -0.4351**
(0.2175)
Controls Yes
Firm Fixed Effects Yes
Year Fixed Effects Yes
SIC Cluster Yes
F-Test (weak instruments) 65.857
N 115,367
57
Table A.2. Capital Expenditure and Initial leverage
This table presents results of analyzing the association between collusion and capital expenditure, using sample splits.
The dependent variable Capital Expenditure is capital expenditure divided by book value of assets. Columns (1) and
(2) present the results of splitting the sample into firms with below-median and above-median Leverage as measured
in the first year of each firm’s observations in the data. Collusion takes a value of 1 for cartel firms during collusion
years, and 0 otherwise; Post Collusion takes a value of 1 for cartel firms during the 5 years after a cartel is dissolved,
and 0 otherwise. The Controls include: Profitability; Tangibility; Cash Flow Volatility; and Sales (see Appendix A
for definitions). Standard errors (in parentheses) are adjusted for heteroscedasticity and clustered at the industry level.
Significant at: *10%, **5% and ***1%.
Dependent variable: Capital Expenditure
Below Median
Leverage
(1)
Above Median
Leverage
(2)
Collusion 0.0290 -0.0153
(0.0256) (0.0184)
Post Collusion 0.0249 -0.0060
(0.0364) (0.0184)
Controls Yes Yes
Firm Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
SIC Cluster Yes Yes
R-squared (within) 0.1926 0.1456
N 56,705 58,436